Give Your AI Agent a Task Board (Decknote's MCP Server)

Your AI agents already generate work: follow-ups, bugs, reminders, the three things it noticed while doing something else. Most of that ends up in a chat log nobody reads twice. Decknote gives an agent a real place to put it — a board your team already has open — through an MCP server. The agent adds tasks, moves and completes them, and searches back through them later. You watch the cards appear live, drag them, and close them. Same board, two kinds of worker.

Why a board and not a log

A log is write-only. The agent notes “check the failing test on Tuesday” and it's gone the moment the conversation scrolls. A board is memory the agent can query: next session it searches “what's still open” and gets the card back, with whatever you added to it in between. That is the whole point of the integration — not another place to dump text, but a shared task list an agent can read as well as write.

Connect it in two minutes

In ChatGPT, Claude, Claude Code or Cursor, add a connector with the address https://decknote.app/mcp. The app opens a Decknote window: log in, click Allow, and you're done — no token to copy. The step-by-step guide has the exact clicks for each app, and the connection shows up in Settings → Connected apps, where you can disconnect it.

For a script or an agent that runs on its own, with nobody there to log in, create an API token in Settings → API Tokens (the free plan includes one) and point the client at /api/v1/mcp with it:

{
  "mcpServers": {
    "decknote": {
      "type": "http",
      "url": "https://decknote.app/api/v1/mcp",
      "headers": { "Authorization": "Bearer ntrl_your_token" }
    }
  }
}

That's it — no separate account for the agent, no webhook to host. Either way the connection carries your permissions: the agent can read any board you belong to and write to any board where you're an editor.

The tools

The agent starts with list_boards to find where things go — or makes its own with create_board and create_list — then works with tasks:

• create_task — add a card to a list.
• search_tasks — the recall tool: find tasks by text, status or due date.
• get_task — read one in full.
• update_task, move_task — edit it, or move it across lists.
• complete_task / reopen_task — mark done or reopen.
• archive_task — clean up; archived cards can be restored.
• add_comment — leave a progress note a human will read.

A typical loop: the agent finishes a job, calls create_task for the three follow-ups it found, and moves on. Tomorrow it calls search_tasksfor what's open, works one, and calls complete_task. Meanwhile you've dragged one to “Doing” and commented on another — and the agent sees both.

What it is and isn't

It's honest about its scope: the agent works with tasks, not with your whole account. It can't change your plan, delete a board, or reach a board you don't have access to — every tool is scoped to your boards with the same permissions as the rest of Decknote. And a board doesn't become an agent-only silo: it's the same real-time board your team uses, which is exactly what makes it useful.

Full setup and the exact request/response shapes are in the API & MCP docs, and there's more on the approach on the Decknote for AI agents page.

Create your free Decknote account and try all of this on your own board — no credit card required.